> Could I ask you what is the reason for using unsmoothed priors for SPM5
> instead of smoothed ones(which were used in previous versions of SPM). Do
> these priors give better answers by experiments? Best regards
The best priors to use are the ones that best reflect the prior probability of
a voxel being of a particular tissue class. In SPM5, the priors should be
overlayed more accurately because the model includes some nonlinear warping.
In SPM2, there was only an affine registration, so overlaying the priors was
less accurate.
There is still a possibility that smoothing the priors may help, but you would
need to attempt some sort of model selection to do this. Will's papers on
model selection would give you some pointers on this. Basically, model
selection provides a framework for properly answering empirical questions
about which model is better. The principles should probably be applied to
models of brain function as well as models for segmenting images.
Best regards,
-John
|